A levy institute model for greece

A LEVY INSTITUTE MODEL FOR GREECE
TECHNICAL PAPER
Dimitri B. Papadimitriou, Gennaro Zezza, and Michalis Nikiforos
May 2013
of Bard College
Levy Economics
Institute
LEVY INSTITUTE
Observatory of Economic and Social Developments, Labour Institute of the Greek General
Confederation of Labour
1

A Levy Institute Model for Greece
Technical Paper
*


Dimitri B. Papadimitriou, Gennaro Zezza, and Michalis Nikiforos

May 2013


* We wish to thank V. Duwicquet for research assistance and Y. Dafermos and M. Nikolaidi for helpful
comments on previous drafts. Any remaining errors are our responsibility.

Copyright © 2013 Levy Economics Institute of Bard College.
2
Table of Contents
Summary 3
1. LIMG: A Levy Institute Model for Greece 3
2. The Data 8

3. Econometric Specification 9
4. Model Properties 27
5. Conclusion 30
Reference 31

3
Summary
In this report, we present the main characteristics of a stock-flow consistent model for Greece,
estimated on quarterly data over the last 30 years and built to provide us with a “tool for thinking,”
in the words of Wynne Godley (the architect of such models), to explore the implications of
alternative policy options for the Greek economy. This report focuses on the technical structure of
the model, which will be used in subsequent Strategic Analysis reports for projecting the trajectories
of the Greek economy for the next four to five years, conditional on assumptions about alternative
economic policies.

1. LIMG: A Levy Institute Model for Greece
In the macroeconomic model for the Greek economy we have developed, henceforth LIMG, we adopt
the “New Cambridge” approach, which is also the basis of the Levy Institute model for the US
economy. That model has shown itself to be extremely effective for constructing reliable medium-term
economic scenarios. In addition, the reconstruction of macroeconomic data for Greece, discussed in an
interim report,
1
has shown many similarities—as well as differences—with the evolution of the US
economy and of other developed economies in Europe. We refer in particular to the stability of private
sector net saving relative to income up to the 1990s, when this flow-flow ratio started to drift toward
negative territory, implying an unsustainable accumulation of private sector debt, which was followed—
in Greece more prominently—by a large and rapidly increasing public sector debt relative to GDP.
The model considers the private sector as a whole, combining households and firms and considering
their receipts and outlays with the other two sectors—the government and the rest of the world—
focusing in particular on their financial balances, which imply in turn a path for the net wealth or debt

of each sector.
The accounting for flows is summarized in Table 1, using the conventional approach of social
accounting matrices pioneered by Richard Stone, where payments are recorded in the columns, receipts
are recorded in the rows, and payments and receipts related to production are dealt with separately.
Transactions that involve a capital account are also considered separately, and could be detailed in a
flow-of-funds matrix.



1 Papadimitriou et al. (2012).
4
Table 1. Social accounting matrix for the LIMG model

1. Prod.
2. Priv. S.
3. Gov.
4. RoW
5. CA
6. Total
1. Production

+PX
+G
+NX

+GDP
2. Private sector
+VAp

+TRgp

+TRwp

+YP
3. Government
+NIT+Ggos
+DT

+TRwg

+YG
4. Rest of the world
+NITw
+TRpw
+TRgw


+YW
5. Capital account

+S
– GDEF
– CA

0
6. Total
+GDP
+YP
+YG
+YW
0

In this first version of the model (as in the Levy Institute US model), we do not consider the stock of
physical capital explicitly, and investment is therefore accounted as part of private expenditure (PX).
This implies that private sector saving (S) is already net of investment spending. In our econometric
analysis, we will nevertheless verify—in future research—our results on the determinants of private
expenditure with an analysis of its main components: consumption and investment.
The matrix records end-of-period values, which are in accounting equilibrium, since—for production—
any discrepancy between production and demand is recorded as a change in inventories treated as
investment and incorporated into private expenditure, and value added is obtained as the residual
between the value of sales and indirect taxes paid to the government.
2

Saving for each sector is obtained as the residual between receipts—the row total—and all other
payments; this therefore ensures that the total of rows two to five is equal to the total for the
corresponding column. The sixth row and column, for “net payments to the capital account,” is
implied by the identities of the previous rows and columns. We use GDEF for the negative saving of
the government and CA for the external current account.
The accounting of Table 1 therefore implies the set of the following identities:
1) GDP = PX + G + NX
2) YP = VAp + TRgp + TRwp – TRpw = GDP – NIT – Ggos – NITw + TRgp + TRwp – TRpw
3) S = YP – PX – DT
4) YG = NIT + Ggos + DT + TRwg
5) GE = G + TRgp + TRgw

2 Profits paid abroad are considered later, as transfers from the private sector to the rest of the world.
5
6) GDEF = GE – YG
7) CA = NX + TRwp + TRwg – (TRpw + TRgw + ITw)
I1) S = GDEF + CA

where the last identity is not included in the numbered sequence because it must be verified by the
previous relations.
We use an oversimplified flow-of-funds matrix, detailed in Table 2.

Table 2. Flow of funds for the model

Priv. S.
Gov.
RoW
Total
Government debt
+ΔDGp
–ΔDG
+ΔDGw
0
Net private sector liabilities
–ΔPSL

+ΔPSL
0
Total
+S = ΔFA
–GDEF
–CA = –ΔFW
0

We are assuming that the government does not hold financial assets, and treat symmetrically private
sector liabilities that are assets of the rest of the world, and financial assets issued by the rest of the
world and held by the private sector. The logic implies that any increase in net saving for the private
sector can only (a) increase the amount of government bills held, and/or (b) increase (decrease) the

amount of foreign (domestic) liabilities held domestically (by the RoW).
The corresponding stocks can be obtained through the standard stock-flow identity:
Stock(t+1) = Stock(t) + Flow(t) + NCG(t) – DS(t)
where NCG stands for net capital gains and DS stands for a reduction in the stock, which may arise, for
instance, by default on a debt. Our experience with the Levy Institute US model shows it is useful to
estimate stocks at historic cost; that is, without taking into account net capital gains. We therefore
constructed the following stock variables, cumulating the flows starting from sensible values at period
“zero”:
8) DG = DG(–1) + GDEF
9) FA = FA(–1) + S
I2) FW = FW(–1) + CA
10) FW = FA – DG
6
where again the value of (the change in) net foreign wealth FW is implied by accumulation of net
financial assets from the private sector and government deficit.
All stocks of wealth or debt will generate interest payments—or profits—in the following periods, and
we are interested in modeling these payments consistently. We are also interested in obtaining more
detail on some payments made by, say, the government sector and the private sector, in order to
endogenize those components—such as unemployment benefits—which vary more or less
automatically in concert with the business cycle.
11) TRpw = Rf·PSL(–1) + NPY + CEPA + TROpw
12) TRgp = Rg·DGp(–1) + SBEN – SOC + TROgp
13) TRgw = Rg·DGw(–1)
where we have made explicit net interest payments from the private sector to the RoW; net property
income (NPY) paid abroad; net compensation of employees paid abroad (CEPA) in total transfers
from the private sector to the RoW; interest paid domestically on government debt; social benefits
(SBEN) and social contributions (SOC) from other net payments from the government to the private
sector; and interest paid abroad on public debt from overall transfers from the government to the
RoW.
Net export (NX) is disaggregated according to

14) NX = XGS – MGS
15) XGS = XG + XS
16) MGS = MG + MS
We next link all variables in euros to quantity indexes, where we append a K to variable names to refer
to “real” variables:
17) PX = ppx·PXK
18) XG = pxg·XGK
19) XS = pxs·XSK
20) MG = pmg·MGK
21) MS = pms·MSK
7
22) G = GCC + GIC + GI
23) GCC = pg·GCCK
24) GIC = pgcf·GICK
25) GI = pg·GIK
26) GK = GCCK + GCIK + GIK
27) YD = YP – DT – Trpw
28) YDK = YD/ppx
29) GPK = GDP/pgdp
where YDK measures real disposable income. Accounting identities for the labor market are
30) POP = LF + RET + NLF + POPY
31) LF = N + U
32) ur = U/LF
This concludes our accounting setup.
In the current specification, the accounting part of the model depends on the following variables:
a) Components of aggregate demand that will be modeled through the econometric specification
discussed below: private expenditure; imports; exports
b) Labor market variables, some of which are determined through econometrics (employment and
unemployment, and hence the labor force; retired people) and some of which will be projected
exogenously (population growth)

c) Prices and relative prices, some of which will be determined through econometrics (import and
export prices), while others will be kept as exogenous, although future model specification
could take them into account
d) Fiscal policy variables: these will be used as policy instruments for conditional forecasts,
although some variables that are only partially under the control of the government, such as
payments for social benefits, will be endogenized appropriately. Other fiscal policy variables
include government expenditure on goods and services—separated into individual and
collective consumption, government investment, current transfers other than interest payments,
8
all ex post implicit tax rates, and net transfers on capital account. Interest payments on the
existing stock of debt are endogenous, given the relevant ex post interest rate. The surplus of
government enterprises is also projected exogenously.
e) Interest rates: at present, we only use the ex post implicit interest rates on public debt and
foreign debt, which are exogenously determined.
f) Determinant of Greek exports: we chose to use the real GDP of Germany as a proxy for the
income variable of the trade partners of Greece, and the deflator of private expenditure in
Germany as the proxy variable to compute competitiveness of Greek exports on foreign
markets.
g) The share of government debt held abroad is exogenously determined.
h) Other variables of less importance are also exogenous, such as other net payments from abroad
other than interest payments or compensation of employees, etc.
2. The Data
As we noted in Papadimitriou et al. (2012), the specification of a model for the Greek economy had to
address severe problems related to the availability of data. Our core dataset is derived from the national
accounts and the nonfinancial sectoral accounts, both published by the Hellenic Statistical Authority
(ElStat from now on), and the sectoral financial accounts published by the Bank of Greece. The
accounts for the real sector are available from 2000 onward, with a statistical break in 2005 making the
data from 2000 to 2004 not comparable to later data, according to ElStat. We have tried to address the
2005 break in the data with appropriate dummy variables in our econometric specification, but they
have (usually) turned out to be of either small or no significance.

We used annual data from the European Commission’s macroeconomic database (AMECO) and the
International Monetary Fund to estimate national accounts backward. This required seasonal
adjustment of the data published by ElStat,
3
quarterly interpolation of the AMECO series prior to
2000, and finally backward estimation, which for some series—such as GDP and its components—is
available back to 1960, while others—mainly government accounts—could only be traced back to
1988. Some series, such as disaggregated trade, could be updated backward more precisely using
Eurostat quarterly data.
Original data have been used to create model variables, yielding strong stock-flow consistency and

3 We are aware of a potential problem in adjusting for seasonality series that contain a break, but non-
seasonally-adjusted data would be more complex to estimate backward. As more reliable data is published by
ElStat, our procedure should produce better-quality data.
9
simplicity, so that, for instance, we simulate real GDP from the sum of its components, thus generating
a small discrepancy on data measured at chained prices, and model the discrepancy so that our
projections can refer to the published figure for real GDP.
Major stocks in the model—that is, the stocks of government debt, net foreign assets, and net financial
wealth of the private sector—are estimated at costs by cumulating the relevant flows. When we then
compare our results with the corresponding variables at market prices, published by the Bank of
Greece for a short time period, the results are satisfactory. They are detailed below.
3. Econometric Specification
As we have shown in Papadimitriou et al. (2012), the Greek economy’s financial balances were
relatively stable up to the 1990s but drifting toward instability at the beginning of 2000. If we had used
only the 2000–12 sample in our econometrics, it would have generated parameters that would
necessarily imply instability for our intermediate-run simulation period. To avoid this problem, we first
eliminated the seasonal components from the ElStat data, and then projected them backward, using
(interpolated) annual data from Eurostat and the AMECO databank.
Components of aggregate demand: Private expenditure

The model is demand driven, with government expenditure exogenously determined by the
government. The core “behavioral” relationship, which is the crucial determinant of the multiplier, is
given by the private expenditure function. Following the New Cambridge tradition, we estimate the
relations between real private expenditure (PXK), real disposable income of the private sector (YDK),
and (the opening value of) the real stock of net financial assets (FA). Should the estimate produce
stable parameter values, they will imply a ratio of financial assets to income toward which the economy
is converging; the (temporary) impact of additional expenditure determinants, such as asset price
bubbles, we capture with additional stationary variables. In symbols,
33) PXK = f(YDK, FA/ppx, other)
In Figure 1, we report the combination of growth in real disposable income and real aggregate
expenditure, which is very strong both in levels and growth rates. The chart shows that in some periods
in the middle of the recession of 2008–09, private expenditure fell more than would be implied by
income, while in other periods, as in 2005–07, it rose more than can be justified by income. This
preliminary analysis confirms the need for additional variables to explain deviations from a stable
expenditure/income norm.
10

Sources: ElStat; authors’ calculations

Note that we adopt a definition of disposable income such that it can be exactly divided between
private expenditure and the net increase of financial assets of the private sector. However, this
definition includes capital transfers that are not always relevant for expenditure. This is the case, for
instance, of a large capital transfer from the Greek government to the banking sector in 2012 to rescue
a failing bank. We tested the effect of such capital transfers on expenditure and, finding no correlation,
decided to use real current disposable income as the determinant of expenditure, and estimated the
change in net financial assets by adjusting, exogenously, for net capital transfers.
To verify our New Cambridge approach, we derived a quarterly estimate of net financial assets of the
private sector, which by accounting definition must equal the sum of public sector debt and net
financial liabilities of the rest of the world. As discussed in Papadimitriou et al. (2012), we obtain these
estimates at costs by cumulating the relevant flows; that is, net saving of the private sector as a whole.

11

Sources: ElStat; Bank of Greece; authors’ calculations
12

Sources: ElStat; Bank of Greece; authors’ calculations

In Figure 2a, we compare our estimate of the change in net government liabilities (over the same
quarter of the previous year) with the identical flow measure published by the Bank of Greece, labeled
“Historic cost – BoG.” The discrepancy between the two measures reflects the fact that national and
sectoral accounts published by ElStat, which are behind our estimate, are not entirely consistent with
the Bank of Greece measure, which uses different sources. However, the two measures at historic cost
have a similar pattern, validating our own process of data reconstruction. The third measure, labeled
“Market price – BoG,” reports the change in the stock of net government liabilities over one year, as
published by the Bank of Greece. This last measure will include changes in the market value of
government liabilities as well as “haircuts,” and will therefore be more volatile and fluctuate around the
historic measure in “normal times.” It has decreased, predictably, with the 2012 “haircut,” which by
construction has no effect on the measure at historic cost.
We have tried alternative estimates of our measure at costs that include the haircut (i.e., including a
one-time decrease in the stock of debt) but the resulting measure does not seem to be consistent with
the published figures for the flows of interest payments. We have therefore preferred to keep a simpler
13
approach, while waiting for longer-term and more reliable data.
In Figure 2b, we report the stocks of net government liabilities implied by the three flow measures
discussed above. The discrepancies between ElStat and BoG flow measures shift at the end of 2003,
leaving a stable gap between the two measures.

Sources: ElStat; Bank of Greece; authors’ calculations
14

Sources: ElStat; Bank of Greece; authors’ calculations

In Figure 3a, we report the flows relative to the second component of the net financial assets of the
private sector—that is, net foreign assets. Again, the discrepancies between the two measures at historic
cost reflect different sources used by ElStat and BoG in constructing their statistics, while the
difference between data at historic cost and market prices in recent years is due to the haircuts on
government debt and fluctuations in the market value of Greek assets held abroad.
By accounting identity, we obtain the stock of net financial assets of the private sector reported in
Figure 4. As noted in Papadimitriou et al. (2012), the measure of net financial assets of the private
sector obtained from ElStat shows that this sector is now in a net debtor position, while the same
measure from BoG data is more optimistic, with financial assets still in positive territory. Comparing
these measures with the stock of government debt in Figure 2b, however, it is quite clear that, even
with the more optimistic measure of financial wealth of the private sector, it is not conceivable to
reduce the stock of government debt by appropriating financial assets from the private sector.
15

Sources: ElStat; Bank of Greece; authors’ calculations

Our econometrics show that the New Cambridge approach, which implies that the link between private
expenditure and disposable income can be expressed as a wealth adjustment function, has empirical
evidence for the Greek economy. However, the economy can wander away from its target wealth-
income norm, even for prolonged periods of time, because of net capital gains on assets or other
changes that boost or depress expenditure relative to income or wealth.
We have thus explored possible links between private expenditure and alternative measures of net
capital gains. In Figure 5, we report two measures of the price of assets that have proven to be effective
indicators for capital gains in our model of the US economy: a price index for the stock market (from
the Bank of Greece, estimated backward with data from the OECD database) and a measure of the
price of existing homes, obtained from the historical series on the prices of dwellings in urban areas
published by the Bank of Greece. Both indexes have been rebased at 100 in 2005. When both price
measures increase in a boom, as in the 1997–99 and 2003–07 periods, they move toward the upper-

right part of the chart. When they both decline, as in the 2007–12 period, they move in the opposite
direction.
16


Sources: Bank of Greece; authors’ calculations

The chart shows, clearly, that the dynamics of these indexes are different over time. A first stock
market boom in the second half of the 1990s saw a moderate increase in the price of housing. With the
2001 recession, speculation moved out of the stock market and into the housing market, which
increased by roughly 46 percent up to 2003, when the stock market regained confidence, and by 102
percent over the 2000–07 period. Housing prices are now back where they were in 2003, while the
stock market is back at its 1996 level (not shown in the chart). The latest data show a continuous
decline in the housing market, while the price index for the stock market has been rising since July
2012. Should the price of housing go back to its 1997 level, as has been the case for the stock market,
home owners would experience a further drop of about 50 percent in the value of their properties,
which, if it materializes, would have further adverse effects on consumption.
Given their respective trend and cycles, the two measures are not correlated. Our econometrics show a
strong correlation between the stock market index and private expenditure, as detailed below, especially
since 2000, while the link between the housing price index and private expenditure is not as robust,
although the variable is significant.
Another potential determinant of private expenditure is the availability of credit. From a theoretical
17
point of view, if potential borrowers are credit rationed, expenditure will be lower than desired for a
given level of current and future expected income. As rationing is lifted, expenditure will increase.
In Figure 6, we report the stock of net financial assets of the household sector, as published by the
Bank of Greece (the household sector includes all noncorporate business, which is relatively large in
Greece). The chart clearly shows the steady increase in the stock of liabilities over the whole period up
to the crisis: the flow of annual borrowing
4

grew from 2 percent of GDP in 1998 to 8.3 percent in the
last quarter of 2007, at the beginning of the current crisis. This trend is, therefore, coherent with the
hypothesis of debt-financed growth in consumption and/or investment from this sector. The spike in
the stock of assets at the end of 1999 is coherent with the data on stock market prices discussed above:
the stock market index went down by more than two-thirds between 1999 (at 182.2) and 2003 (at 52),
and this largely explains the collapse in the value of financial assets held by households. A better
understanding of these trends would require additional data on the distribution of wealth, since it is
probable—if patterns are similar to those in other developed countries—that the owners of financial
assets are highly concentrated in the top decile of the income distribution; as is known, however, most
income deciles, as well as the vast majority of small businesses, borrow.

4 Measured as the change in the stock of debt over the same quarter of the previous year.
18

Sources: ElStat; Bank of Greece

In Figure 7, we report the same measures for corporate nonfinancial business. The overall trend in total
liabilities for this sector is dominated by the change in the value of equities, but if we consider only the
stock of loans outstanding,
5
annual borrowing fluctuated around 7 percent of GDP from 1998 to 2006
and started to increase more rapidly only in 2006, presumably because other sources of funds had dried
up. Borrowing at a rate of 7 percent of GDP is not necessarily problematic when nominal GDP growth
is around 8 percent, as was reported to be the case in Greece—on average—in the period up to the end
of 2006. Borrowing of corporate businesses became unsustainable only when the recession started in
2007.

5 Measured as the sum of short- and long-term securities plus short- and long-term loans from Table 3.1L of
the financial accounts published by the Bank of Greece.
19

Sources: ElStat; Bank of Greece

To complete our analysis, in Figure 8 we report the same measures for the financial sector. Note the
large increase in the stock of liabilities that starts in 2004: this is mainly due to the increase in the value
of equities issued by the Greek financial sector, which were once held mainly by the Greek household
sector (62 percent at the end of 1997) and since then increasingly held abroad. The share of equities
held by foreigners increased from 12 percent at the end of 1997 to 35 percent at the end of 2006,
before the recession started, and presumably acted as part of the collateral for increased borrowing
from abroad.
20

Sources: ElStat; Bank of Greece

Our preferred equation for determining private expenditure, estimated over 1988Q1–2012Q3, is the
following (Table 3):
6








6 The equation has passed Engle-Granger and Philip-Ouliaris tests for cointegration available in Eviews. The
final equation has been determined through a general-to-specific approach, and nonsignificant variables are
not reported. Stars indicate significance at 1 percent (***), 5 percent (**), or 10 percent (*). We have not been
able so far to test for weak exogeneity of real disposable income, since we have not yet found a good set of
instruments.

21
Table 3. Private expenditure function
Determinant
Impact
Long-run
Real disposable income
0.34***
0.85***
Real net financial assets
0.01*
0.02*
Household borrowing
0.230***
0.587***
Stock market price
8.658**
22.069**
Price of housing
3.222*
8.213*

The equation shows that an increase of 100 euros in real disposable income implies an increase of 34
euros in private expenditure in the same quarter, and of 85 euros overall. Coefficients for real net
financial assets and (real) borrowing can be read in a similar way, while the stock market price and the
housing price are indexes, and therefore coefficients measure how much private expenditure would rise
(in 2005 million euros) with an increase of one point in the index.
Trade
Exports are broken down between goods and services. In both cases they are determined by a proxy
for the income of foreigners—which for convenience we chose to be real GDP in Germany—and by
relative prices, where again we use the deflator of private expenditure in Germany as the basis to

compute Greek competitiveness in foreign markets.
7

Exports of goods and services—at current prices—were a small share of Greek GDP up to the
beginning of floating exchange rates in the 1970s, when they grew steadily from about 10 percent of
GDP to around 20 percent, which is the average for the 1980–2006 period. With the current crisis, they
dropped by 25 percent from the peak in 2008 and a trough at the end of 2009, and then started to
recover. A more in-depth analysis of Greek trade has been provided in Papadimitriou et al. (2012); we
just want to recall here that Greek exports have been increasing moderately against a continuous
devaluation of the drachma against the deutsche mark, up to the beginning of the euro era.
When addressing the econometric specification of exports, we chose to break down the aggregate into

7 A measure of the income and inflation rate of major Greek foreign partners can be computed in future
improvements on the model, but it will raise the cost of maintaining the model, as it will add several variables
that need to be updated and projected for simulation.
22
goods and services that are likely to have different determinants. In the aggregate, some measures of
price competitiveness are reported in Figure 9, which shows that, even after the crisis, inflation in
Greece—as measured by the deflator of private expenditure—has been somewhat larger than in
Germany, a trend that reversed only in 2012. Prices of Greek exports apparently rose even quicker than
domestic prices.


Sources: Eurostat; ElStat; authors’ calculations

When looking at the same measures separately for goods and services, we noticed a structural break in
the data, possibly related to the construction of national accounts, which are not strictly comparable
before and after 2005. In Figure 10, we report the same measures of competitiveness, now related only
to goods. Notice the sudden jump in the price of exports relative to both domestic prices abroad and
Greek imports, while no such jump occurs in the price of Greek imports relative to domestic prices.

23

Sources: Eurostat; ElStat; authors’ calculations

A jump in the opposite direction, though less pronounced, is visible in the same price measures, now
related to the export and import of services reported in Figure 11.
24

Sources: Eurostat; ElStat; authors’ calculations

Notice that the price of exports of services seems to be more stable than the price of goods, when
compared to measures of foreign competitors.
Given the analysis above, we used a dummy variable to test the hypothesis that the relation between
exports and relative prices could shift in 2005. Indeed, we find a large price elasticity in this last part of
our sample.
Our preferred equation for exports of goods has the following properties (Table 4):